Hi Giuseppe,
One point is that to estimate smoothness based on the residuals,
those residuals should basically be white noise with some amount
of smoothness.
The residuals from the individual regression fit this, in theory.
And since there are so many TRs to estimate this over, it becomes
a very robust estimate, especially further averaged over subjects.
However, there is little reason (that I can think of) to expect
that the stdev values behave as white noise. And note that they
are not even residuals (deviations from some expected value).
It seems feasible to get residuals from some group analysis, but
obtaining the robustness of those from single subject analysis
seems difficult.
----
So separate from most of this is a question of where to get
white residuals. It seems possible to get them from the group
analysis, but I see no benefit. To me, this should not depend
on any model, except that we do our best to remove any known
signal in the data.
If you get a residual dataset by removing all known means and
correlates from the group data, that might be one volume, or
even a bunch. But from all EPI volumes of all subjects, that
might be 30000 volumes to average over. It is very robust.
- rick